DFU-VGG, a Novel and Improved VGG-19 Network for Diabetic Foot Ulcer Classification

Abstract

A complication caused by diabetes mellitus is the appearance of lesions in the foot region called Diabetic Foot Ulcers (DFU). Delayed treatment can lead to infection or ulcer ischemia, leading to lower limb amputation in an advanced stage. This article proposes the DFU-VGG, a convolutional neural network (CNN) inspired by convolutional blocks of VGG-19 but with smaller dense layers and batch normalizations operations. To specify the DFU-VGG parameters, we fine-tuned s even different CNN architectures using two image datasets containing 8,250 images with different color, contrast, resolution, and texture features. The proposed evaluation identifies f our c lasses: none, ischemia, infection, and both. Our approach achieved 93.45% of accuracy and an excellent Kappa index of 89.24%.